Split expressions into static and dynamic

[FFroehlich]

static expressions don't depend on time/states and only need to be evaluated after reinitialization, not at every timepoint.

[dweindl]

Also, some expressions only need to be evaluated at measurement timepoints. Could also treat them separately, or just flatten everything in y and remove unused expressions afterwards.

[dweindl]

If we split the expressions as suggested, how would we handle them in ReturnData? Combine static+dynamic expressions a stick to the current ReturnData::w? I.e., the separation of static/dynamic would be hidden to the user.

[dweindl]

How to implement?

A: void Model::fw(realtype t, realtype const* x) -> void Model::fw(realtype t, realtype const* x, bool static_=true, bool dynamic=true)
body: if (static_) { /* static exprs */ } if (dynamic) { /* dynamic exprs */ }

vs

B: add: void Model::fw_static(realtype t, realtype const* x), void Model::fw_dynamic(realtype t, realtype const* x), void Model::fw(realtype t, realtype const* x) { fw_static(t, x); fw_dynamic(t, x);}

vs

C: as B but with additional split of w values into w_static + w_dynamic

I think I'd go for A as this minimizes the required code changes.

[FFroehlich]

Yeah A sounds pretty straightforward. Would go for a single flag though as I don't think we ever want to evaluate static only.

[dweindl]

Not sure if the initial issue was about w or model expressions in general. Just to be explicit: It might make sense to do the same for other model functions as well.

[FFroehlich]

initial issue was only about w. what other expressions could this be applied to?

[dweindl]

initial issue was only about w. what other expressions could this be applied to?

Depending on the model, there are quite some time-independent entries for the various derivatives. But not sure whether the benefit outweighs the additional complexity.

[FFroehlich]

initial issue was only about w. what other expressions could this be applied to?

Depending on the model, there are quite some time-independent entries for the various derivatives. But not sure whether the benefit outweighs the additional complexity.

ah, well , haven't looked at the PR yet, but it would be great to do the same for dwdw, dwdx and dwdp.

[dweindl]

Right, but potentially also beyond dwd*.

[dweindl]

ah, well , haven't looked at the PR yet, but it would be great to do the same for dwdw, dwdx and dwdp.

I'll look into that separately, after review/merge of #2303.